cure-lab / SCINet

The GitHub repository for the paper: “Time Series is a Special Sequence: Forecasting with Sample Convolution and Interaction“. (NeurIPS 2022)
Apache License 2.0
633 stars 128 forks source link

Do I get better result? #6

Closed EasonHee closed 2 years ago

EasonHee commented 2 years ago

Hi all,

I I changed code in block 1 to code in block 2, the result get better. Is there anything wrong? On dataset(ETTh1, ETTh2, ETTm1) /block1/ (SCINet.py line 107-114)

        x_even = x_even.permute(0, 2, 1)
        x_odd = x_odd.permute(0, 2, 1)

        d = x_odd.mul(torch.exp(self.phi(x_even)))
        c = x_even.mul(torch.exp(self.psi(x_odd)))

        x_even_update = c + self.U(d)
        x_odd_update = d - self.P(c)

/block1/

/block2/

        x_even = x_even.permute(0, 2, 1)
        x_odd = x_odd.permute(0, 2, 1)

        d = x_even
        c = x_odd

        x_even_update = c
        x_odd_update = d

/block2/ Thanks.

Eason